Sandbox Agent

Example Use Cases

• 1

  Secure, isolated code execution environments with persistent state.

• 2

  Multi-runtime support including Python, Node.js, and R.

• 3

  Integration with MCP-compatible clients and REST APIs for sandboxed code execution.

Description

Sandbox Agent runs each code execution session in a hardened, isolated Docker container supporting Python, Node.js, and R runtimes. It offers persistent kernel state, crash detection, and rich output capture, enabling safe and efficient code execution for language model agents. The MCP server exposes tools for session management and code execution, integrating seamlessly with clients like Cursor and Claude Desktop, and supports cross-session file transfer and REST API access.